Alex Sherman

The through-line

Sherman's central claim is that every five to seven years the consumer internet throws up a new marketing channel, and that AI is the current one. Search, social, mobile all followed the same script: a technology changes how consumers use the internet, marketers initially have no idea what to make of it, consumer adoption scales, and eventually the channel exists whether brands like it or not [2][3]. What makes AI different is speed and scope. The legacy channels "took three or four years to compose and reach kind of cruising altitude. AI has reached cruising altitude in 12 months" [2][3], and it is not sitting politely alongside the others: "it was like a channel that was eating all the other channels and becoming the sort of meta channel" [1].

He tracks his own shift in reception as evidence. Two years ago, pitching CMOs on AI as the next big consumer channel meant pointing at ChatGPT and Perplexity when they were, by marketer standards, small, in the tens of millions of users [1][2][3]. It went "from a novel idea kind of pie in the sky" to something search teams experimented with to a whole-organisation priority, and he now claims that "if you were to call any CMO on the Fortune 500, this this is top three priority um at this point" [1]. Bluefish sits in that gap: a platform giving marketing teams real-time analytics on whether and how ChatGPT, Google Gemini and AI mode, Anthropic's Claude and Amazon Rufus describe and recommend their products, plus tools to influence those outcomes [2][3][6].

On why AI search is not SEO with a new coat of paint

The difference, he insists, is the output, and everything else follows from it. A search algorithm ends in ten blue links, so a brand's only question is where it ranks in that list and whether the consumer clicks through [1]. An LLM ends in "a long- form qualitative response to that consumer prompt that is that is rich and immersive and has detail and is personalized and individualized" [1]. That makes the model far hungrier than a ranking algorithm: to sustain a detailed conversation about your products it needs much richer inputs about use cases, benefits and the range of prompts it might receive [1]. His sharpest observation about the resulting awkwardness for marketers is that "websites were not really designed to be data delivery mechanism for large large language models" [1]. They were built for humans, and to some extent for search crawlers. So the work is more content, more dynamic content, and content deliberately built to teach models about the brand [1][2][3].

The questions brands bring him are correspondingly different from rank tracking. Search teams, content, brand marketing, PR and comms all want to know "how are the major models describing us to those consumers? Are we showing up? Are we showing up positively, negatively? How are we compared to our competitors?" [2][3] Bluefish's job as he frames it is diagnostic before it is prescriptive: figuring out what is driving the models, how they are learning about a brand, and what is informing the way they characterise its products [2][3].

On why gaming the models is a dead end

He is blunt that the current moment contains a lot of loophole-hunting, teams "trying in some cases to find loopholes like hey maybe we can create some content that can sort of trick you know open AI and and you know give us a temporary boost," and he calls that "incredibly shortsighted and something that is inevitably going to go away" [1]. The argument is structural rather than moral. Anything that misleads how ChatGPT answers a consumer erodes that consumer's trust in ChatGPT, so the providers have every incentive to close the gap [1]. What follows is "a flight to quality higher quality content," which he reads as good news for content teams, since nobody is especially excited to produce low-quality work [1]. The explicit line in the sand: more content and richer content does not mean AI slop [1].

On what consumers actually did

His read on adoption is that consumers were not persuaded, they were retrained. AI "completely collapses that funnel" of manual keyword research across a dozen sites, and "allows you to get to truth faster" [2][3]. The shopping version is a prompt describing yourself, your priorities and your budget and getting back a ranked comparison with strengths and weaknesses in ten seconds [2][3]. He pushes back on the assumption that this is an early-adopter phenomenon: older generations have moved quickly because they never liked the keyword-and-blue-links internet in the first place, speak in natural language, and do not want to click through a dozen sites and a wall of ads to get an answer [2][3]. He points to research that two thirds of frequent online shoppers already use AI assistants for purchasing decisions [1]. And it is not confined to one vertical, he sees it across financial services, CPG and apparel [2][3].

On the incumbents' math problem

Sherman explicitly refuses the death-of-search framing, calling Google "a phenomenal company" with the talent to work this out, as Gemini shows [2][3]. But he identifies a specific arithmetic bind: Google's ad business was predicated on keyword search and ten blue links, and if the answer now sits in a summary at the top, then "how do we make with one what we were previously making with 10?" [2][3] The messy Google SERP of the past year, with AI overviews above and blue links below, he reads as the visible residue of that experiment, and as a company that was forced into AI by an existential threat rather than choosing the timing [2][3].

Amazon has the same problem in retail form. Its results page is heavily monetised with retail media units, and that high-margin ad business subsidises infrastructure, logistics, shipping and low prices, so "it is not a nice to have business. It is the ex existential kind of driver of their margin" [2][3]. He expects Rufus to move from a small chat box in the corner to the centrepiece of the experience, with ads already appearing inside it [2][3]. Citing the Walmart CTO's stated intent to replace the entire search engine with an AI agent, and noting drily that "Walmart has a habit of saying things like that uh, you know, 3 years before they actually do something about it," he still expects every large retailer on the same journey: meet the shopper in AI, then make sure the retail media revenue follows them there [2][3]. That is also his explanation for platforms blocking third-party agents. They are on the back foot, trying to keep access to their own customer base, keep a role in the shopper journey, and keep the transaction on their own site [2][3]. His overall verdict is that AI is "a reshuffleling of the deck," with every big tech company reconsidering whether it plays in ads, search, enterprise or commerce, and "the rules are being rewritten as we speak" [2][3].

On the first AI holiday season

He treats the Black Friday through Cyber Monday stretch as a live referendum rather than a data point. Consumers spend a trillion dollars in the holiday season and hundreds of billions in marketing budget is deployed to reach them, and for the big retailers those days were "basically like the Super Bowl" [2][3]. The problem is that nobody knows what an AI-powered holiday season means: how people shop for gifts through ChatGPT, what it does to Rufus, whether the transaction closes inside AI or on a retailer site [2][3]. And the budgets chasing those shoppers "were built and designed in a previous version of the internet," planned six or twelve months earlier [2][3].

On leading through turbulence

The metaphors he reaches for are pirate ships and aircraft, and both do the same work. Startups "really should not exist," they are unnatural, under-resourced, swimming upstream against incumbents whose systems you are asking customers to rip out [1]. He treats underdog status as an asset, liberating in what you build, how you build it and who you hire, and he will fly a literal pirate flag in the office to energise the team [1]. He values people who join early-stage companies for the reflex of asking "why not why can't we go do this?" [1]. The aviation version is about temperament under pressure: "are you comfortable flying the plane through the storm," meaning are you running around screaming or do you have situational awareness and your bearings [1]. He argues this matters commercially, not just culturally, because Bluefish's actual product is helping the largest companies in the world navigate a period of genuine uncertainty, and a customer in that position needs an even hand [1]. Nearly twenty years in startups is what he credits for the calm, a shared trait with his two co-founders, who are also repeat entrepreneurs [1].

The corollary is a discipline about signal. He rejects the idea that the industry's changes are coherent or cumulative: "there's so much noise in this category. There's a lot of there's a lot of contradictory things that are happening at the same time," and the leadership job is sifting the meaningful changes from the hype of the day [1]. Alongside that, he argues that building enterprise software has to be approached as a service business anchored in what customers actually need rather than in the technology alone [1].

On the outsider's advantage and how founders get made

He connects his own entrepreneurial instincts to not fitting in. Growing up half-French outside New York meant not belonging neatly to any tribe, and he generalises the pattern: look at most founders' pasts and "there's always a moment where they didn't quite fit in" [1]. The practical residue is fluency across worlds, "a lot of like code switching," which is what he does daily moving between AI scientists, large commercial brands who need that technical work translated, and employees who need the vision [1]. He also thinks the industry has grown into this, becoming more cosmopolitan and international than when he started in 2009, with startup hubs across Australia, Asia and Europe and more languages spoken in San Francisco than ever, which he sees as straightforwardly good for the ecosystem [1].

His entry into startups was accidental. He graduated in 2008 as a proud history major ready for a conventional corporate path that the financial crisis had erased, and a friend suggested a startup while he looked for "a real job" [1]. He fell for it immediately: "I was 22 by by any, you know, by any stretch, I should not have been given any, you know, any responsibilities. But at a startup, you can be 22 and have lots of responsibilities" [1]. What hooked him was a small team attacking a big problem with no sense of inhibitions or blockers, creative, intense and meritocratic [1]. He still reads history for the same reason he reads markets, on the view that there is a great deal to learn about the present by looking at previous ages [1].

Takeaways

• The consumer internet produces a new marketing channel every five to seven years, and AI is the current one, except it reached cruising altitude in twelve months rather than three or four years [2][3].
• Optimising for LLMs differs from SEO because the output is a conversation rather than a ranked list, which makes models far more content and data hungry than search algorithms [1].
• Websites were built for humans and search crawlers, not as data delivery mechanisms for large language models, which is the core structural problem marketing teams now face [1].
• Attempts to trick models into favourable answers are shortsighted, because misleading responses erode consumer trust in the AI provider, whose incentive is to close the loophole; expect a flight to higher-quality content [1].
• Google's bind is arithmetic: an ad business built on ten blue links must now earn from one summary what it previously earned from ten results [2][3].
• Amazon's retail media revenue is the existential driver of its margin, so Rufus will move from side chat box to centrepiece, with ads following the shopper into it [2][3].
• Older consumers are adopting AI search fast because they never liked the keyword-and-blue-links internet, so this is not just an early-adopter shift [2][3].
• Holiday marketing budgets deployed in the first AI-powered season were planned six to twelve months earlier for a previous version of the internet [2][3].
• The leadership test in a noisy category is staying even-handed under turbulence and separating meaningful change from the hype of the day [1].